AGI

Marc Andreessen, asked on Lenny Rachitsky's podcast in February 2026 to define AGI, gave three answers: an AI capable of producing research equivalent to a doctoral thesis, the point where an AI takes control of its own development and improves itself, and the pragmatic one, that we will know when we get there. Needing three tries is the concept's defining feature. AGI is the declared objective of the large AI labs, systems substitutable for humans, with no settled test for having arrived.

Luc Julia, co-creator of Siri, rejects the premise. His book «L'IA n'existe pas» argues that current systems perform sophisticated pattern recognition rather than intelligence, that they neither reason nor understand nor genuinely learn, and that AGI is improbable with current approaches, perhaps impossible. He attributes the hype to commercial interests more than to technical reality. «Arvind Narayanan» holds a similarly opposed position.

Ethan Mollick, in July 2024, treats the disagreement as a planning problem. A 2023 survey of computer scientists put the median AGI date at 2047, with a ten percent probability before 2027, while AI leaders announce timelines of 2027 or five years. Mollick's response is scenario planning across four futures: plateau, linear growth, exponential growth, AGI. He invokes the Moore's law precedent, where a technological expectation became a motivating target and then a self-fulfilling prophecy. That precedent leaves the open question: whether AGI names a destination, or organizes an industry's behavior toward one.